4 * Copyright (c) 2003-2008 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/timer.h"
28 #include "sysemu/replay.h"
29 #include "sysemu/cpus.h"
39 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
40 #include <sys/prctl.h>
43 /***********************************************************/
46 typedef struct QEMUClock {
47 /* We rely on BQL to protect the timerlists */
48 QLIST_HEAD(, QEMUTimerList) timerlists;
56 QEMUTimerListGroup main_loop_tlg;
57 static QEMUClock qemu_clocks[QEMU_CLOCK_MAX];
59 /* A QEMUTimerList is a list of timers attached to a clock. More
60 * than one QEMUTimerList can be attached to each clock, for instance
61 * used by different AioContexts / threads. Each clock also has
62 * a list of the QEMUTimerLists associated with it, in order that
63 * reenabling the clock can call all the notifiers.
66 struct QEMUTimerList {
68 QemuMutex active_timers_lock;
69 QEMUTimer *active_timers;
70 QLIST_ENTRY(QEMUTimerList) list;
71 QEMUTimerListNotifyCB *notify_cb;
74 /* lightweight method to mark the end of timerlist's running */
75 QemuEvent timers_done_ev;
80 * @type: type of clock
82 * Translate a clock type into a pointer to QEMUClock object.
84 * Returns: a pointer to the QEMUClock object
86 static inline QEMUClock *qemu_clock_ptr(QEMUClockType type)
88 return &qemu_clocks[type];
91 static bool timer_expired_ns(QEMUTimer *timer_head, int64_t current_time)
93 return timer_head && (timer_head->expire_time <= current_time);
96 QEMUTimerList *timerlist_new(QEMUClockType type,
97 QEMUTimerListNotifyCB *cb,
100 QEMUTimerList *timer_list;
101 QEMUClock *clock = qemu_clock_ptr(type);
103 timer_list = g_malloc0(sizeof(QEMUTimerList));
104 qemu_event_init(&timer_list->timers_done_ev, true);
105 timer_list->clock = clock;
106 timer_list->notify_cb = cb;
107 timer_list->notify_opaque = opaque;
108 qemu_mutex_init(&timer_list->active_timers_lock);
109 QLIST_INSERT_HEAD(&clock->timerlists, timer_list, list);
113 void timerlist_free(QEMUTimerList *timer_list)
115 assert(!timerlist_has_timers(timer_list));
116 if (timer_list->clock) {
117 QLIST_REMOVE(timer_list, list);
119 qemu_mutex_destroy(&timer_list->active_timers_lock);
123 static void qemu_clock_init(QEMUClockType type, QEMUTimerListNotifyCB *notify_cb)
125 QEMUClock *clock = qemu_clock_ptr(type);
127 /* Assert that the clock of type TYPE has not been initialized yet. */
128 assert(main_loop_tlg.tl[type] == NULL);
131 clock->enabled = (type == QEMU_CLOCK_VIRTUAL ? false : true);
132 clock->last = INT64_MIN;
133 QLIST_INIT(&clock->timerlists);
134 main_loop_tlg.tl[type] = timerlist_new(type, notify_cb, NULL);
137 bool qemu_clock_use_for_deadline(QEMUClockType type)
139 return !(use_icount && (type == QEMU_CLOCK_VIRTUAL));
142 void qemu_clock_notify(QEMUClockType type)
144 QEMUTimerList *timer_list;
145 QEMUClock *clock = qemu_clock_ptr(type);
146 QLIST_FOREACH(timer_list, &clock->timerlists, list) {
147 timerlist_notify(timer_list);
151 /* Disabling the clock will wait for related timerlists to stop
152 * executing qemu_run_timers. Thus, this functions should not
153 * be used from the callback of a timer that is based on @clock.
154 * Doing so would cause a deadlock.
156 * Caller should hold BQL.
158 void qemu_clock_enable(QEMUClockType type, bool enabled)
160 QEMUClock *clock = qemu_clock_ptr(type);
162 bool old = clock->enabled;
163 clock->enabled = enabled;
164 if (enabled && !old) {
165 qemu_clock_notify(type);
166 } else if (!enabled && old) {
167 QLIST_FOREACH(tl, &clock->timerlists, list) {
168 qemu_event_wait(&tl->timers_done_ev);
173 bool timerlist_has_timers(QEMUTimerList *timer_list)
175 return !!atomic_read(&timer_list->active_timers);
178 bool qemu_clock_has_timers(QEMUClockType type)
180 return timerlist_has_timers(
181 main_loop_tlg.tl[type]);
184 bool timerlist_expired(QEMUTimerList *timer_list)
188 if (!atomic_read(&timer_list->active_timers)) {
192 qemu_mutex_lock(&timer_list->active_timers_lock);
193 if (!timer_list->active_timers) {
194 qemu_mutex_unlock(&timer_list->active_timers_lock);
197 expire_time = timer_list->active_timers->expire_time;
198 qemu_mutex_unlock(&timer_list->active_timers_lock);
200 return expire_time <= qemu_clock_get_ns(timer_list->clock->type);
203 bool qemu_clock_expired(QEMUClockType type)
205 return timerlist_expired(
206 main_loop_tlg.tl[type]);
210 * As above, but return -1 for no deadline, and do not cap to 2^32
211 * as we know the result is always positive.
214 int64_t timerlist_deadline_ns(QEMUTimerList *timer_list)
219 if (!atomic_read(&timer_list->active_timers)) {
223 if (!timer_list->clock->enabled) {
227 /* The active timers list may be modified before the caller uses our return
228 * value but ->notify_cb() is called when the deadline changes. Therefore
229 * the caller should notice the change and there is no race condition.
231 qemu_mutex_lock(&timer_list->active_timers_lock);
232 if (!timer_list->active_timers) {
233 qemu_mutex_unlock(&timer_list->active_timers_lock);
236 expire_time = timer_list->active_timers->expire_time;
237 qemu_mutex_unlock(&timer_list->active_timers_lock);
239 delta = expire_time - qemu_clock_get_ns(timer_list->clock->type);
248 /* Calculate the soonest deadline across all timerlists attached
249 * to the clock. This is used for the icount timeout so we
250 * ignore whether or not the clock should be used in deadline
253 int64_t qemu_clock_deadline_ns_all(QEMUClockType type)
255 int64_t deadline = -1;
256 QEMUTimerList *timer_list;
257 QEMUClock *clock = qemu_clock_ptr(type);
258 QLIST_FOREACH(timer_list, &clock->timerlists, list) {
259 deadline = qemu_soonest_timeout(deadline,
260 timerlist_deadline_ns(timer_list));
265 QEMUClockType timerlist_get_clock(QEMUTimerList *timer_list)
267 return timer_list->clock->type;
270 QEMUTimerList *qemu_clock_get_main_loop_timerlist(QEMUClockType type)
272 return main_loop_tlg.tl[type];
275 void timerlist_notify(QEMUTimerList *timer_list)
277 if (timer_list->notify_cb) {
278 timer_list->notify_cb(timer_list->notify_opaque, timer_list->clock->type);
284 /* Transition function to convert a nanosecond timeout to ms
285 * This is used where a system does not support ppoll
287 int qemu_timeout_ns_to_ms(int64_t ns)
298 /* Always round up, because it's better to wait too long than to wait too
299 * little and effectively busy-wait
301 ms = DIV_ROUND_UP(ns, SCALE_MS);
303 /* To avoid overflow problems, limit this to 2^31, i.e. approx 25 days */
304 if (ms > (int64_t) INT32_MAX) {
312 /* qemu implementation of g_poll which uses a nanosecond timeout but is
313 * otherwise identical to g_poll
315 int qemu_poll_ns(GPollFD *fds, guint nfds, int64_t timeout)
319 return ppoll((struct pollfd *)fds, nfds, NULL, NULL);
322 int64_t tvsec = timeout / 1000000000LL;
323 /* Avoid possibly overflowing and specifying a negative number of
324 * seconds, which would turn a very long timeout into a busy-wait.
326 if (tvsec > (int64_t)INT32_MAX) {
330 ts.tv_nsec = timeout % 1000000000LL;
331 return ppoll((struct pollfd *)fds, nfds, &ts, NULL);
334 return g_poll(fds, nfds, qemu_timeout_ns_to_ms(timeout));
339 void timer_init_full(QEMUTimer *ts,
340 QEMUTimerListGroup *timer_list_group, QEMUClockType type,
341 int scale, int attributes,
342 QEMUTimerCB *cb, void *opaque)
344 if (!timer_list_group) {
345 timer_list_group = &main_loop_tlg;
347 ts->timer_list = timer_list_group->tl[type];
351 ts->attributes = attributes;
352 ts->expire_time = -1;
355 void timer_deinit(QEMUTimer *ts)
357 assert(ts->expire_time == -1);
358 ts->timer_list = NULL;
361 static void timer_del_locked(QEMUTimerList *timer_list, QEMUTimer *ts)
365 ts->expire_time = -1;
366 pt = &timer_list->active_timers;
372 atomic_set(pt, t->next);
379 static bool timer_mod_ns_locked(QEMUTimerList *timer_list,
380 QEMUTimer *ts, int64_t expire_time)
384 /* add the timer in the sorted list */
385 pt = &timer_list->active_timers;
388 if (!timer_expired_ns(t, expire_time)) {
393 ts->expire_time = MAX(expire_time, 0);
397 return pt == &timer_list->active_timers;
400 static void timerlist_rearm(QEMUTimerList *timer_list)
402 /* Interrupt execution to force deadline recalculation. */
403 if (timer_list->clock->type == QEMU_CLOCK_VIRTUAL) {
404 qemu_start_warp_timer();
406 timerlist_notify(timer_list);
409 /* stop a timer, but do not dealloc it */
410 void timer_del(QEMUTimer *ts)
412 QEMUTimerList *timer_list = ts->timer_list;
415 qemu_mutex_lock(&timer_list->active_timers_lock);
416 timer_del_locked(timer_list, ts);
417 qemu_mutex_unlock(&timer_list->active_timers_lock);
421 /* modify the current timer so that it will be fired when current_time
422 >= expire_time. The corresponding callback will be called. */
423 void timer_mod_ns(QEMUTimer *ts, int64_t expire_time)
425 QEMUTimerList *timer_list = ts->timer_list;
428 qemu_mutex_lock(&timer_list->active_timers_lock);
429 timer_del_locked(timer_list, ts);
430 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
431 qemu_mutex_unlock(&timer_list->active_timers_lock);
434 timerlist_rearm(timer_list);
438 /* modify the current timer so that it will be fired when current_time
439 >= expire_time or the current deadline, whichever comes earlier.
440 The corresponding callback will be called. */
441 void timer_mod_anticipate_ns(QEMUTimer *ts, int64_t expire_time)
443 QEMUTimerList *timer_list = ts->timer_list;
446 qemu_mutex_lock(&timer_list->active_timers_lock);
447 if (ts->expire_time == -1 || ts->expire_time > expire_time) {
448 if (ts->expire_time != -1) {
449 timer_del_locked(timer_list, ts);
451 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
455 qemu_mutex_unlock(&timer_list->active_timers_lock);
458 timerlist_rearm(timer_list);
462 void timer_mod(QEMUTimer *ts, int64_t expire_time)
464 timer_mod_ns(ts, expire_time * ts->scale);
467 void timer_mod_anticipate(QEMUTimer *ts, int64_t expire_time)
469 timer_mod_anticipate_ns(ts, expire_time * ts->scale);
472 bool timer_pending(QEMUTimer *ts)
474 return ts->expire_time >= 0;
477 bool timer_expired(QEMUTimer *timer_head, int64_t current_time)
479 return timer_expired_ns(timer_head, current_time * timer_head->scale);
482 bool timerlist_run_timers(QEMUTimerList *timer_list)
485 int64_t current_time;
486 bool progress = false;
489 bool need_replay_checkpoint = false;
491 if (!atomic_read(&timer_list->active_timers)) {
495 qemu_event_reset(&timer_list->timers_done_ev);
496 if (!timer_list->clock->enabled) {
500 switch (timer_list->clock->type) {
501 case QEMU_CLOCK_REALTIME:
504 case QEMU_CLOCK_VIRTUAL:
505 if (replay_mode != REPLAY_MODE_NONE) {
506 /* Checkpoint for virtual clock is redundant in cases where
507 * it's being triggered with only non-EXTERNAL timers, because
508 * these timers don't change guest state directly.
509 * Since it has conditional dependence on specific timers, it is
510 * subject to race conditions and requires special handling.
513 need_replay_checkpoint = true;
516 case QEMU_CLOCK_HOST:
517 if (!replay_checkpoint(CHECKPOINT_CLOCK_HOST)) {
521 case QEMU_CLOCK_VIRTUAL_RT:
522 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL_RT)) {
529 * Extract expired timers from active timers list and and process them.
531 * In rr mode we need "filtered" checkpointing for virtual clock. The
532 * checkpoint must be recorded/replayed before processing any non-EXTERNAL timer,
533 * and that must only be done once since the clock value stays the same. Because
534 * non-EXTERNAL timers may appear in the timers list while it being processed,
535 * the checkpoint can be issued at a time until no timers are left and we are
538 current_time = qemu_clock_get_ns(timer_list->clock->type);
539 qemu_mutex_lock(&timer_list->active_timers_lock);
540 while ((ts = timer_list->active_timers)) {
541 if (!timer_expired_ns(ts, current_time)) {
542 /* No expired timers left. The checkpoint can be skipped
543 * if no timers fired or they were all external.
547 if (need_replay_checkpoint
548 && !(ts->attributes & QEMU_TIMER_ATTR_EXTERNAL)) {
549 /* once we got here, checkpoint clock only once */
550 need_replay_checkpoint = false;
551 qemu_mutex_unlock(&timer_list->active_timers_lock);
552 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL)) {
555 qemu_mutex_lock(&timer_list->active_timers_lock);
556 /* The lock was released; start over again in case the list was
562 /* remove timer from the list before calling the callback */
563 timer_list->active_timers = ts->next;
565 ts->expire_time = -1;
569 /* run the callback (the timer list can be modified) */
570 qemu_mutex_unlock(&timer_list->active_timers_lock);
572 qemu_mutex_lock(&timer_list->active_timers_lock);
576 qemu_mutex_unlock(&timer_list->active_timers_lock);
579 qemu_event_set(&timer_list->timers_done_ev);
583 bool qemu_clock_run_timers(QEMUClockType type)
585 return timerlist_run_timers(main_loop_tlg.tl[type]);
588 void timerlistgroup_init(QEMUTimerListGroup *tlg,
589 QEMUTimerListNotifyCB *cb, void *opaque)
592 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
593 tlg->tl[type] = timerlist_new(type, cb, opaque);
597 void timerlistgroup_deinit(QEMUTimerListGroup *tlg)
600 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
601 timerlist_free(tlg->tl[type]);
605 bool timerlistgroup_run_timers(QEMUTimerListGroup *tlg)
608 bool progress = false;
609 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
610 progress |= timerlist_run_timers(tlg->tl[type]);
615 int64_t timerlistgroup_deadline_ns(QEMUTimerListGroup *tlg)
617 int64_t deadline = -1;
619 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
620 if (qemu_clock_use_for_deadline(type)) {
621 deadline = qemu_soonest_timeout(deadline,
622 timerlist_deadline_ns(tlg->tl[type]));
628 int64_t qemu_clock_get_ns(QEMUClockType type)
631 QEMUClock *clock = qemu_clock_ptr(type);
634 case QEMU_CLOCK_REALTIME:
637 case QEMU_CLOCK_VIRTUAL:
639 return cpu_get_icount();
641 return cpu_get_clock();
643 case QEMU_CLOCK_HOST:
644 now = REPLAY_CLOCK(REPLAY_CLOCK_HOST, get_clock_realtime());
647 case QEMU_CLOCK_VIRTUAL_RT:
648 return REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, cpu_get_clock());
652 uint64_t qemu_clock_get_last(QEMUClockType type)
654 QEMUClock *clock = qemu_clock_ptr(type);
658 void qemu_clock_set_last(QEMUClockType type, uint64_t last)
660 QEMUClock *clock = qemu_clock_ptr(type);
664 void init_clocks(QEMUTimerListNotifyCB *notify_cb)
667 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
668 qemu_clock_init(type, notify_cb);
671 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
672 prctl(PR_SET_TIMERSLACK, 1, 0, 0, 0);
676 uint64_t timer_expire_time_ns(QEMUTimer *ts)
678 return timer_pending(ts) ? ts->expire_time : -1;
681 bool qemu_clock_run_all_timers(void)
683 bool progress = false;
686 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
687 if (qemu_clock_use_for_deadline(type)) {
688 progress |= qemu_clock_run_timers(type);